BDW17 London - Totte Harinen, Uber - Why Big Data Didn’t End Causal Inference
Simulation and collectiveintelligence
1. Simulation and Collective
Intelligence
Gloria Origgi
CNRS –InstitutNicod
Workshop on Simulation, Paris ENS,
June 16-18 2011
2. What are Simulations?
Computer simulations are a vast and
heterogeneous class, which contains
processes as varied as that of flying a plane,
the movement of people in public transport
systems, network-based computations,
strategic scenarios of warfare (kriegspiele) and
virtual reality scenarios of all kind
Hartmann, 1996
6. Crop Predictions
The Iowa agriculture landscape: Green areas are more productive for
soy, corn, and wheat; red are least. (Lanworth Company)
7. Different ways of conceptualizing
simulations (Hartmann)
① Simulations as a technique: Investigate the
detailed dynamics of a system
② Simulations as a heuristic tool: Develop
hypotheses, models and theories
③ Simulations as a substitute for an experiment:
Perform numerical experiments
④ Simulations as a tool for experimentalists:
Support experiments
⑤ Simulations as a pedagogical tool: Gain
understanding of a process
8. A Fundamental Difference with
Models:
No need for elegance and simplicity: Data
crunching by computers is so powerful that
we can complexify the input data in order to
obtain a better empirical adequacy.
9. Simulations as Building Scenarios
• Scenarios are a political technique
• Three requirements:
– Credibility
– Legitimacy
– Salience
10. Haas/Stevens 2011: .
• Credibilitymeans that the knowledge claims are believed to
be accurate: within a consensus theory of truth that means
that they are publicly created through a deliberative
process by people widely regarded as experts.
• Legitimacymeans that it is developed by people who have
social authority, and that it is accepted by people outside
the community that developed it. In practice this often
rests on peer review and scholarly reputation.
• Saliencemeans that the information is timely, and is
organized on a politically meaningful time scale and scale of
resolution”
11.
12. My point:
• Computer simulations are prediction-driven. If you cannot control
the accuracy of the calculation, you can control the correctness of
prediction.
• In this sense, the are a special class of models that bears a strong
relation with policy-making and political/economical institutions.
• The questions simulations are supposed to answer aren’t always
questions that raise within the scientific community, rather, they
are normative charged societal questions that are put forward by a
political agenda.
• This raises a problem of prediction and focus. Many events have
been not predicted because of a blindness of the models, that were
able to answer only the top-down questions put on the table by
policy makers.
13.
14. Computer Simulation: The End of
Theory?
Chris Anderson: “The quest for knowledge used to
begin with grand theories. Now it begins with
massive amounts of data. Welcome to the
Petabyte Age.” Petabytes allow us to say:
"Correlation is enough." We can stop looking for
models. We can analyze the data without
hypotheses about what it might show. We can
throw the numbers into the biggest computing
clusters the world has ever seen and let statistical
algorithms find patterns where science cannot.
15. Example: Google
• Google spell checker doesn’t have any model
of language or grammar, just correlates the
number of “yes” triggered by a certain
spelling.
• Google translator
• Craig Venter’s sequencing of entire
ecosystems.